Automatic Detection of Tropical Cyclones from Satellite Images Using YOLO Models

Tropical cyclone (TC) movement shows a significant impact on both human lives and infrastructure. TCs are considered as one of the most destructive natural disasters that can cause severe casualties and economic losses. The importance of accurate detection of TCs is emphasized for preventing and mitigating associated natural disasters. To address these issues, a real-time and cost-effective framework for efficiently detecting cyclones from satellite images is proposed in this study. Various detection models, including YOLOv3, Tiny-YOLOv4, and YOLOv5, are introduced for identifying TCs. Experimental results indicate that the proposed YOLOv5 model notably outperforms state-of-the-art models in terms of detection accuracy. Compared with YOLOv3 and Tiny-YOLOv4, the highest mean average precision of 90.80% and detection speed of 70 f/s are achieved by YOLOv5, respectively. This advancement lays the groundwork for developing a high-speed cyclone detection tool to mitigate damages caused by cyclone disasters.